Performance Comparison of SRCNN, VDSR, and SRDenseNet Deep Learning Models in Embedded Autonomous Driving Platforms

Autonomous driving is one of the most popular research areas. The camera is the eyes of autonomous driving. The computer vision technology using such a camera is a core element for the popularization of autonomous driving. Autonomous driving is not only improving the quality of the images but also processing real-time images. In the last few years, the performance of vision tasks have been largely improved via the deep-learning-based approaches. Despite of this advantages, deep-learning-based approaches have a limitation in that resulting in expensive computation. In this context, we analyze these approaches in the point of the processing time on the embedded autonomous driving platform, NVIDIA TX2. We used the three representative deep-learning-based super-resolution methods, SRCNN, VDSR, and SRDenseNet for evaluation. In the evaluation, the comparison of processing speed over different resolutions was conducted.